Techniques to provide improved wind input for operating offshore wind turbines

A method for operating a component of a wind turbine farm comprising a plurality of wind turbines, a control center, or a service platform, the method comprising:

• setting a geographic area of interest that encompasses the plurality of wind turbines in the wind turbine farm to be operated;
• setting a forecast time interval from a selectable current time to a forecast horizon time of interest;
• setting a maximum lag time interval for using mesoscale weather forecasts to describe wind conditions up to the horizon time of interest,
• collecting mesoscale weather forecasts from a weather model for a machine learning (ML) training time interval TT at a plurality of model grid locations of interest that encompasses the geographic area of interest, wherein the duration of the ML training time interval TT is at least ten times the maximum lag time interval;
• collecting fine-scale wind measurements at locations of two or more of the plurality of wind turbines during the ML training time interval TT;
• determining selected parameters of the mesoscale weather forecasts and coefficients of an evolving ML forecast model based on the mesoscale weather forecasts and the fine-scale wind measurements during the ML training time interval TT ending at a current time;
• using mesoscale weather forecast values for the selected parameters during a lag time interval up to the maximum lag time interval, and the coefficients of the evolving ML forecast model, to produce a forecast wind at each location of the plurality of wind turbines at a future time during the forecast time interval after the current time;
• and
• operating the component of the wind turbine farm based on the forecast wind at the location of at least one of the plurality of wind turbines.